Using Deep Learning and Machine Learning Methods to Diagnose Hailstorms in Large-Scale Thermodynamic Environments

Farha Pulukool, Longzhuang Li, Chuntao Liu

Sustainability · 15 December 2020

In one sentence. Random forest, autoencoder and CNN models diagnose hailstorm occurrence globally from four ERA5 thermodynamic attributes, with random forest performing best.

Abstract

Hailstorms have caused damages in billions of dollars to industrial, electronic, and mechanical properties such as automobiles, buildings, roads, and aircrafts, as well as life threats to crop and cattle populations, due to their hazardous nature. Hence, the relevance of predicting hailstorms in the future has significant scientific, economic, and societal benefits. However, climate models do not have adequate resolutions to explicitly resolve these subscale phenomena. One solution is to estimate the probability of these storms by using large-scale atmospheric thermodynamic environment variables from climate model outputs, but the existing methods only carried out experiments on small datasets limited to a region, country, or location and a large number of input features. Using one year of Tropical Rainfall Measuring Mission (TRMM) observations and European Center for Medium-Range Weather Forecasts (ECMWF) Re-Analysis Interim (ERA-Interim) reanalysis on a global scale, this paper develops two deep-learning-based models (an autoencoder and convolutional neural network (CNN)) as well as a machine learning approach (random forest) for hailstorm prediction by using only four attributes—convective potential energy, convective inhibition, 1–3 km wind shear, and warm cloud depth. In the experiments, the random forest approach produces the best hailstorm prediction performance compared to the other two methods.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.